Group 3 (2024-25) Democratizing AI for MSMEs with Small E-commerce Language Models

Micro, Small, and Medium Enterprises (MSMEs), crucial to many economies, face significant barriers in adopting advanced AI technologies due to the high computational and financial costs of large-scale models.
This limits their ability to engage with AI-driven innovations.
To address this, our project focuses on developing and fine-tuning Small Language Models (SLMs) specifically for MSMEs.
Our study empirically demonstrates that SLMs, finetuned on domain-specific datasets, achieve comparable or superior performance to general-purpose LLMs. In support of this stance, we train and test some powerful SLMs across various e-commerce tasks, including sentiment analysis, product recommendation, and attribute extraction as a PoC.
In addition to diminished inference costs via SLMs, we also employ Parameter Efficient Fine Tuning via QLoRA to further reduce the training costs for creating domain specific chat models. The resultant models consistently outperform larger models like GPT-4 Turbo and Gemini Pro in domain specific evaluations. By demonstrating that SLMs can deliver competitive results with minimal resources, our research contributes to the broader discourse on AI efficiency, advocating for lightweight, specialized models that balance performance with accessibility.

Presented By- Sayali Kawatkar, Ritesh Bhalerao, Aum Kulkarni, Dyotak Kachare
Mentor- Sangeeta Oswal

Видео Group 3 (2024-25) Democratizing AI for MSMEs with Small E-commerce Language Models канала VESIT AI&DS